yev0-4b
A 4B System One decision model from XY Space: a LoRA on Qwen3.5-4B-Base, merged, that answers choice, yes/no (noul) and score questions about a state with a calibrated probability per option, read from the option-letter logits in one forward pass. Apache-2.0.
Built by Cho Yin Yong of XY Space. The "Stage 0" model: one epoch of LoRA (rank 64) on 44k rows, merged at the root with the adapter alongside; the prompt format is Together's Tev1, used verbatim. Takes 2 to 6 options and reads letters A to F, with per-type temperatures. English only; accuracy drops beyond about 4k tokens; weak on date arithmetic and long multi-step workflows. XY Space's own runs elsewhere: JevBench public items 78.4% (a third-party board) and R-Judge at chance. DecideBench is XY Space's own benchmark. Synthetic labels were written and checked by Claude Opus 5.5, with no human review. Serving code (yev serve, /v1/systemone) is on GitHub; no hosted API.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one probability
At a glance
| Parameters | 4B |
| Base model | Qwen/Qwen3.5-4B-Base |
| Maker | XY Space |
| Released | 2026-10-04 |
| License | apache-2.0 |
| Reported accuracy | 94.5% |
Get the weights
pip install systemonemodels
systemone pull xy-space/yev
The files are served from the maker's Hugging Face repository, choyiny/yev0-4b, and verified against the checksums recorded here.
Read more
This page was opened by System One for XY Space, who can claim the organisation and take it over at any time.